Carlos P Mendoza

AI • Finance • Risk • Health

Research Projects

Hedging Targeted Risks with Reinforcement Learning
ASTIN Bulletin (submitted, 2025)
What: Developed a reinforcement learning framework to hedge life insurance guarantees.
How: Combined adaptive AI decision-making with Shapley explainability and deep statistical analysis.
Impact: Reduced global risk exposure more effectively than traditional hedging, while avoiding unwanted risk transfers.

Deep Hedging with Options using the Implied Volatility Surface
Management Science (submitted, 2025)
What: Addressed a multidimensional portfolio hedging problem where AI agents manage several financial instruments simultaneously.
How: Designed reinforcement learning models that incorporate the full implied volatility surface as input to guide hedging strategies across multiple assets.
Impact: Delivered more robust portfolio hedges and superior performance compared to traditional multi-asset risk management approaches.

Is Deep Hedging vs Delta Hedging a Statistical Arbitrage?
Finance Research Letters (2024)
What: Compared AI-driven hedging strategies with classical delta hedging.
How: Applied backtesting and rigorous statistical arbitrage tests.
Impact: Established the conditions under which AI methods truly outperform without introducing arbitrage or market distortions.

Enhancing Deep Hedging of Options with Implied Volatility Surface Feedback Information
Mathematical Finance (submitted, 2024)
What: Investigated whether deep reinforcement learning agents can exploit volatility surface information in a univariate hedging task (single option).
How: Introduced volatility feedback signals into the agent’s decision-making and analyzed their effect on performance.
Impact: Demonstrated that volatility-driven features improve hedge accuracy, confirming their informational value even in single-instrument settings.

Parameter estimation for partially observed stable continuous-state branching processes
Working Paper (2025)
What: Developed new statistical methods for estimating parameters of complex stochastic processes.
How: Combined Laplace inversion techniques with probabilistic modeling.
Impact: Created robust tools for analyzing systemic risk and other phenomena under uncertainty.